use numpy sinc

[argriffing]

No description provided.

[coveralls]

Coverage remained the same when pulling 7cc24c1 on argriffing:remove-sinc into 371b4ff on scipy:master.

[rgommers]

+1, no need to duplicate this function.

I checked the numpy implementation, and that looks fine to me. The 1e-20 used there doesn't affect the precision it seems.

[rgommers]

The tests for special.sinc should also be removed in this PR.

EDIT: maybe keep a very basic one to check that the function still exists in the special namespace.

[coveralls]

Coverage remained the same when pulling 284023e on argriffing:remove-sinc into 371b4ff on scipy:master.

[argriffing]

@rgommers I removed most of the testing.

[rgommers]

Thanks, merging.

[endolith]

Would it make sense to move the tests and implementation into numpy instead of just deleting them? scipy's implementation makes more sense to me than something that depends on float inaccuracy.

[argriffing]

@endolith The numpy sinc passed the scipy tests, and numpy has its own more extensive sinc testing. On the other hand I'd like a new C sinc ufunc implementation in either numpy or scipy.

[rgommers]

New ufunc would indeed be nice. That should go into numpy and replace its current implementation imho.

[argriffing]

New ufunc would indeed be nice. That should go into numpy and replace its current implementation imho.

related: numpy/numpy#7322

[madphysicist]

For the immediate future, numpy/numpy#7322 will actually remove the possibility of making np.sinc into a ufunc. However, I also have plans to work around that.